從3D + 1D的前胸超音波影像中自動找尋左心室

Abstract

3D+1D的心臟超音波影像是近年來在醫學攝影技術中新崛起的一項技術,它秉持了超音波無害、方便、且便宜的優點, 更進一步得到更完整的心臟資訊。但由於其資料量十分龐大,使得單純以人工處理的作業變得十分繁瑣。 在本篇論文中, 我們提出的方法可省去人工處理所耗費的大量的人力及物力, 並且得到正確的左心房心室分割線及左心房的位置。方法分為三個步驟, 第一步先利用2-means algorithms決定僧帽瓣底部的點, 這些點以左心房為分界兩兩相連可行成一些線。在這些線中以optical flow、combinatorial optimization決定每張影像中左心房與左心室分割線的位置。第二步則是在分割線的位置之上找尋左心室。我們利用α-shapes來找尋每一張影像上的框再以combinatorial optimization在這些框中決定正確的左心室。最後由 snake models將左心室輪廓最佳化。在本論文中, 為了避免超音波影像中雜訊的干擾, 每一張影像都先行由SML filter 過濾雜訊。我們以兩組不同的資料來實作這個方法, 並且證實了可得到正確的結

We present an automatic method to determine the Left Ventricle (LV) from 3D+1D Transthoracic Echocardiograms with very few user inputs. There are three steps in the proposed method and every image is reduced the speckle noises by a SML filter before processed. First, we find the sequences of mitral annular lines. The optical flow technique and a 2-means algorithm are used to determine a set of possible mitral annular points, and a pair of mitral annular points could form the mitral annular line. The problem to determine the sequence of mitral annular lines in 2D + 1D echocardiograms is modeled as a problem to find the shortest path in a graph. Then, α-shapes are used to segment all 2D echocardiograms, and the results are many contours including LV in one image. We reused the combinatorial optimization technique to determine the sequence of left ventricles in 3D + 1D echocardiograms i.e. the shortest path in the graph. The last step is to optimize the LV contours by the snake model. Experimental results show that we can find accurate mitral annular lines and left ventricles. Our goal to reconstruct the 3D+1D model of Left Ventricle is to evaluate the cardiac function such as the ejection fraction and wall motion.